Hi list, I have been trying to understand NFS's working - focusing mainly on the way I/O (write back) is performed. But after a lot of tries I am stuck at some very stupid issues which I am unable to sort out. I would really appreciate if someone can give a pointer or two. Please correct me wherever my understanding is not right. Also, I had posted this same message on linux.nfs list but I suppose people there were qutie busy and I did not get any response (or possibly these are very stupid questions) - so posting here in hope that someone might be able to help me out. I apologize for the length of this mail. ----------------------------------------- I am using stock 2.6.16.21 kernel for understanding purpose (and NOT focusing on NFS v4) Problem: 1) As per my understanding NFS_INO_INVALID_DATA flag means that if it is set, flush the data which is in the write back list. For writing, nfs_file_write calls nfs_revalidate_mapping function. Which in turn calls nfs_sync_mapping if it is a regular file in question and if INO_INVALID_DATA flag is set. nfs_sync_mapping calls nfs_wb_all which would flush back all the pages to be written back (for any pervious I/O). Is it that: a) if invalid data is set means that data should be flushed back - always. b) invalidating data (with flag) does *not* mean that it should NOT be : written back and page_cache_release(d). <-- is there any such condition (situation) in which such a step has to be taken (assuming NON SYNC operations). 2) I created a module which would scan the list of pages associated with an Inode. I ran this for an I/O done over NFS for checking page flags like: Uptodate, Dirty, Writeback, Buffered and Private. My observation was: 2.1) Only Buffered and private bits are on for any I/O (unflushed i.e. not executed using SYNC). 2.2) All the pages had there Uptodate flag set - and no page was NOT(uptodate). 2.3) Same was the case for Dirty flag of the page. Hence, is it that: a) Pages used, refered by NFS are *always* Uptodate. Hence, kernel would be able to read pages without flushing them (without the knowledge of NFS). If the above is true, THEN: --> Assuming a read calls comes for a file which had pending write --> backs. Read call is for same pages which were to be written back (and are in cache with Uptodate flag ON). --> Then, kernel would simply read all the pages which are uptodate and --> copy them back to the user space. (do_generic_mapping_read function).HOW does NFS handle such a request? (it never gets the control of the read request which is returned from the kernel itself) OR, it is OK to read from pages which are in writeback list of pages? Also, I could infer that: nfs_readpage function would NEVER be called for any page which is uptodate (in cache) even if has to be written back. Still there is a check for pending writes (nfs_wb_all) in this function which would be redundant. Regards Shreyansh -- To unsubscribe from this list: send an email with "unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx Please read the FAQ at http://kernelnewbies.org/FAQ
